摘要
PVM的设计目标是连接一个异构的计算机和网络的集合,给用户提供一个并行计算平台来运行他们的MPP或集群计算应用。运行在集群上的应用经常碰到诸如网络故障、主机故障和进程故障等各种异常故障,可靠性是这类平台面临的主要问题。PVM能够应付主机和网络故障并且提供轮询机制和故障通知原语来建立容错应用。由于在设计PVM时对在异构环境下软件的适用性投入了太多的考虑,PVM原有的故障检测机制不能满足实时性的要求。为了解决这个问题,我们提出并实现了一种改进PVM检错机制实时性能的方法,使得对主机故障事件的发现时间缩短了五千多倍。
PVM is designed to link to a heterogeneous collection of computers and network and provide users with a parallel platform for running their MPP or duster computer applications. Applications running in dusters are easily exposed to problems like network failure, node failure and process failure. Dependability is a major issue in this type of platforms. Fortunately, PVM is able to withstand host and network failures and provide polling and notification primitives to allow fault - tolerant applications to be built. Because the design of PVM pays so much attention to the software＇ s applicability in heterogeneous environment, PVM＇ s fault - detection mechanism isn＇t able to meet real- time requirement. For solving this problem, it implementes a method of improvement for real - time fault - detection mechanism in PVM which the time that is spent in detecting host crash decreases 5000 times than before improvement.
出处
《计算机与数字工程》
2006年第2期1-4,11,共5页
Computer & Digital Engineering
基金
船舶基金(项目编号:41901080101)
关键词
PVM
集群
故障检测
PVM, cluster, fault - detection
